Skip to content

v0.3.2 — LLM Search Intelligence + README Rewrite

Choose a tag to compare

View all tags
@zychenpeng zychenpeng released this 27 Feb 13:48
v0.3.2
30233ce

What's New

LLM-Powered Search Intelligence (Render API)

  • Claude Haiku 4.5 generates optimal search queries from idea descriptions in any language
  • Automatic fallback to dictionary pipeline on any failure
  • New POST /api/extract-keywords endpoint (rate-limited: 50/IP/day)
  • keyword_source field in response meta — shows whether "llm" or "dictionary" was used

Architecture

  • MCP stdio: dictionary-only pipeline (100% anchor hit on 54-idea golden set, no external dependency)
  • Render API: LLM-first with dictionary fallback (better Chinese/non-English support)

README Rewrite

  • New positioning: "We search. They guess."
  • Competitor comparison table vs ChatGPT/ValidatorAI/IdeaProof
  • "Why not just ask ChatGPT?" section
  • Agent auto-trigger instructions

Infrastructure

  • MCP Registry metadata (server.json) prepared
  • Agent instructions template (examples/agent-instructions.md)
  • 120 tests passing (18 new)
  • LLM vs dictionary eval report on 54 golden ideas

Full changelog: v0.3.1...v0.3.2

Assets 2
Loading